Research On The Problem Of Personalized Recommendation Based On Mobile Users

With the acceleration of the informatization process and the advent of the 5G era,domestic operators are actively carrying out technological upgrades and product innovations.As one of the three major domestic operators,China Mobile accurately grasps the development trend of the industry and cooperates with major platforms in the entertainment,e-commerce and other fields to provide mobile users with an abounded of business products.In the process of the transformation of business products from the single voice,traffic,SMS,etc.to diversification,the drawbacks of extensive marketing have become apparent.Therefore,it has become the focus of mobile companies that how to fully mine users’information and realize personalized recommendation aiming to them,and precise marketing.At present,the research results of personalized recommendation have been applied to all aspects of life,but hardly in the field of mobile communication.In order to provide an idea and method of personalized recommendation for mobile users,this paper studies the classic matrix factorization recommendation model in the recommendation system.The main contents and conclusions of the research are as follows:First,because the users with similarity in terms of basic in-formation and communication behavior will share analogously products,we propose a new matrix factorization model in the recommendation system including the basic information and communication behavior of the mobile users on account of the fac-tors affecting the product order.It turns out that the improved model proposed in this paper has better performance than traditional models on the problem of rating prediction.On the Top-N recommendation problem,the improved matrix factor-ization model is more precise,but has a slight decrease and increase in coverage and average popularity respectively.Second,in the rating prediction,because of the difficulty of direct explicit rating data collection and the subjective rating behavior,this paper adjusts the calculation methods and meanings of some indicators in the RFM model combined with the characteristics of mobile users in the product order-ing process and constructs a user-product rating matrix by the RFM model,which can more accurately and objectively reflect the preference of users for products,and then applies the results to the rating prediction.Third,in the Top-N recommenda-tion,adopting a negative sampling method that prioritizes popular products that users have not ordered into the negative sample,and using the improved matrix factorization model to calculate the interest of use.The Top-N recommendation sequences obtained by the improved matrix factorization can provide a choice for the personalized recommendation of mobile users.